Published in Chemistry Europe: Ultra-low Data Screening With Active Learning

Our April preprint on ultra-low data drug candidate screening has now been published in Chemistry Europe.

The paper demonstrates that active learning with as few as 110 affinity evaluations can reliably recover top-1% hits from large compound libraries — making high-throughput screening tractable for resource-limited academic settings. The best-performing strategy combined continuous and data-driven descriptors (CDDD) with a multi-layer perceptron (MLP) and pairwise difference regression (PADRE) data augmentation, achieving a 97% probability of finding at least five top hits.